MI-BILSTM prediction method considering feature importance value fluctuation

The invention provides an MI-BILSTM prediction method considering feature importance value fluctuation, and belongs to the field of short-term power load prediction in a power system. The method comprises the following steps: firstly, extracting importance values of input characteristics at different moments in advance to form an importance value fluctuation matrix; secondly, the original input features are dynamically corrected through a matrix, so that fluctuation information is contained in the original input features, that is, importance value fluctuation of the features is extracted through a mutual information method, the original input features are dynamically corrected, and the information of the importance value fluctuation is fused in the corrected input features; and finally, the corrected input features are substituted into the BILSTM network for short-term load prediction. The method not only retains the advantages of a weight sharing structure in the aspect of parameter simplification, but also